Fix a table generation bug for pme-shift

Fix a table generation bug for pme-shift

The shift correction was not applied correctly to PME and PME-user
tables, but a value of 0 was used instead. This will not have
influenced any normal runs since the shift correction is applied
directly (analytically) in the current kernels. In theory it
might have resulted in 10^-5 relative errors at the cutoff that
are only likely to be visible in double precision if you were
(1) using PME-user electrostatics with custom tables, and
(2) applying shift modifiers (switch is fine). This is not a
setup we are testing anyway, and the PME-user options are likely
to go soon.

Change-Id: Ibd4d2388713879a6613938929edc2cba142b7c9b